基于SVM的灌区无人机影像渠系提取

Extraction of Irrigation Networks of UAV Orthophotos Based on SVM Classification Method

灌区渠系制图配合现代节水灌溉技术,对合理配水、安全输水有着重大影响。但目前普遍使用的灌区遥感影像分辨率不高,给渠系提取与制图带来一定的困难。本文将无人机采集的高精度正射影像、高程、坡度数据相结合作为数据源,提取出具有显著描述能力的渠系特征来构建训练样本集,基于支持向量机的分类方法对目标渠系进行分割提取,再通过后处理对提取结果进行去噪、连接和优化,实现了无人机高分辨率多数据源的渠系提取。结果表明,该渠系提取方法可以识别提取灌区中的支渠、斗渠和部分农渠,渠系连续性良好,与手绘渠系对比,精度最高可达89.35%。其中提取误差主要由级别较低的渠系中渠床淤泥沉积导致影像、地形特征不明显造成。

Irrigation district canal system with modern water-saving irrigation technology has a significant impact on rational distribution of water and the safety of water supply. However,the resolution of the commonly used remote sensing image of irrigation area is not high,which brings difficulties to the extraction and mapping of the drainage system. The high-precision ortho-image,elevation and slope data collected by UAVs were taken together as data sources. Features with strong canal discriminative ability were obtained from them to construct a training set. The classification system was trained via the support vector machine to segment canals from images. Then,the extraction results were denoised,connected and optimized,and the canal extraction of UAV high resolution multi-source data was realized. The results showed that the canal extraction method can identify the branch canal in the irrigation area.Meanwhile,competitive performance was achieved in the continuity of the canal,the bucket and the part of canal system. The precision was up to 89. 35%. The extraction error was mainly caused by the deposition of canopy mud in the lower canal system which made the terrain features not easy to be recognized. In conclusion,the method proposed provided a new solution for the extraction of irrigation and drainage canal and can be applied to the actual agricultural production.